Text Region Localization and Recognition for ID Card Identification using Deep Learning Approaches

Abstract

Deep Learning approaches are currently successful in the application areas of object detection and recognition. In Optical Character Recognition (OCR) applications, there are still challenges in the problems of text region localization and segmentation. Various conventional methods that used in OCR systems still couldn’t get high accuracy and still to do to be more effective for real time OCR applications. This paper presented recognition and identification of the specified identity document (ID) to solve those problems by applying deep learning techniques. For the process of text region extraction, the Regional Proposal Networks (RPN) such as Faster R-CNN is used to identify which areas are the most probable of text in the ID card. For text sequence learning, Recurrent Neural Network (RNN) learns timestamp-based text sequences from images to avoid segmentation problems and recognize a given text. According to the experimental results, the deep learning techniques give a satisfied accuracy and loss rates in text region identification and recognition for ID cards.